JP2003061099A - Motion detection method in encoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a motion detection method for preventing scattering of motion vectors and minimizing the encoding quantity. SOLUTION: The method includes a step S2 of discriminating, to which of an edge part, a flat part and a detail part a macro block type expressing the property of an image of macro blocks of a video signal belongs, a step S4 of calculating an estimated amount g of an error power code amount of a macro block and an estimated amount h of a motion vector code amount, with respect to each motion vector MV object of the macro block, in the case of the edge part or the flat part, a step S5 of selecting a motion vector MV, where the sum of the estimated amount g of the error power code amount of the macro block and the estimated amount h of the motion vector code amount will become minimum, or a step S8 of selecting a motion vector MV for minimizing a SAD (sum of absolute differences) of the macro block in the case of the detail part.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion detecting method in a coding device, and more particularly to a motion detecting method suitable for use in a motion compensation system for video signal coding.

[0002]

2. Description of the Related Art Conventionally, a method called block matching is generally used as a motion vector detecting method used for motion compensation in moving image coding.
In this method, it is determined how much each block of the input image currently being encoded moves with respect to the prediction image of one frame before, by calculating the absolute value of the difference in pixel value between the macroblock in the vicinity of the prediction image. Sum of values (called "SAD")
Etc., and the vector to the macroblock with the smallest value is used as the motion vector.

[0003]

However, when the motion vector is obtained by the above-mentioned conventional method, there is a high possibility that there are a plurality of motion vectors having the minimum SAD, especially in a flat image area. Therefore, conventionally, there has been a problem that the motion vector is scattered and the code amount is increased.

An object of the present invention is to solve the above-mentioned problems of the prior art, to provide a motion detecting method capable of preventing motion vector scattering and minimizing the code amount.

[0005]

In order to achieve the above-mentioned object, the present invention determines a macroblock type representing the image property of a macroblock of a video signal, and the macroblock type is a flat area of the image. In this case, the feature is that the motion vector that minimizes the sum of the estimated amount of error power code amount of the macroblock and the estimated amount of code amount of motion vector is selected.

According to this feature, it is possible to prevent the scattering of the motion vector in the flat area of the image and minimize the code amount thereof.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of an embodiment of the present invention. In the following, MPEG-
2, TM5 will be described as an example, but the present invention is not limited to this.

When the input image data a is input, the macroblock extraction unit 1 extracts a macroblock of 16 × 16 pixels from the input image data a, and the subtractor 2, the macroblock type determination unit 13 and It is sent to the motion detector 12. The subtractor 2 receives the macroblock and the block data (prediction image) of the motion compensated previous frame from the prediction image creating unit 11, and generates an inter-frame difference signal (prediction error signal). The prediction error signal is supplied to the orthogonal transformation unit 3 and subjected to orthogonal transformation such as DCT. The switching unit 14 is connected to the contact point c in the I picture mode, and is connected to the contact point b in other cases. The bit rate of the orthogonal transform coefficient obtained by the orthogonal transform unit 3 is reduced by the quantizing unit 4 and is sent to the variable length coding unit 5 and the inverse quantizing unit 6.

The variable length coding unit 5 assigns codes to the output of the quantization unit 4 so that the average code length becomes short. On the other hand, the inverse quantization unit 6 inversely quantizes the output of the quantization unit 4 and supplies the output to the inverse orthogonal transformation unit 7, and the inverse orthogonal transformation unit 7
Outputs the output of the inverse quantization unit 6 to the adder 8 after performing inverse orthogonal transform.

The adder 8 adds the motion-compensated block data of the previous frame from the prediction image holding unit 11 and the decoded difference data from the inverse orthogonal transform unit 7, and the frame memory 9
Send to. The frame memory 9 delays the input block data by one frame and supplies it to the motion compensation unit 10 and the motion vector detection unit 12 as previous frame data. The frame memory 9 can be used as the predicted image holding unit 11.

The macroblock type determination unit 13 determines the type of macroblock output from the macroblock extraction unit 1. That is, it is determined whether the macroblock is the detail part of the original image, the flat part which is a region where there is almost no change in brightness, or the edge part which includes only a single straight edge. To do. This determination result is sent to the motion vector detection unit 12. As will be described later, the motion vector detection unit 12 detects a motion vector of a block having a small code amount from the input image data a and the previous frame data according to the determination result, and outputs it to the motion compensation unit 10. The block data from the frame memory 9 is supplied to the motion compensating unit 10. The motion compensating unit 10 motion-compensates the block data with a motion vector to create motion-compensated previous frame block data. . The motion-compensated previous frame block data is supplied to the subtractor 2 via the prediction image holding unit 11.

Next, the function and operation of the motion vector detecting section 12, which is the main part of the present invention, will be described with reference to the flowchart of FIG.

First, the principle of the present invention will be described. In general, when a motion compensation prediction error is obtained in an image block including only a straight line edge or a region where there is almost no change in luminance, S
There is a high possibility that a plurality of motion vectors MV that minimize AD will appear. In such a region, it is considered that the motion vector MV estimation often involves a large error. Therefore, in the present invention, it is determined whether the macro block MB for motion estimation has a property of the detail part, the edge part, or the flat part, and the block having the property of the edge part or the flat part is determined. If it is determined that the motion vector MV is obtained by the method of the present invention described below, the conventional block matching method is not used. On the other hand, when it is determined that the block has the property of the detail portion, the motion vector MV obtained by the conventional block matching method is regarded as reliable, and the motion vector MV obtained by the block matching method is adopted as usual. To do.

In step S1 of FIG. 2, the block number i of the macroblock MB is set to 1. In step S2,
The motion vector detection unit 12 determines the macroblock (MB) type based on the output from the macroblock type determination unit 13. When it is determined by the determination that it is the edge portion or the flat portion, the process proceeds to step S3, and the motion vector MV candidate of the macroblock i is set. Next, in step S4, an error power code amount estimate g and a motion vector MV code amount estimate h are calculated for each motion vector MV candidate.

Here, the estimation g of the error power code amount is given.
Is the prediction error variance of each block _i ^Two, Consider the visual characteristics
Using the quantised scale code mquant,
It can be expressed as in equation (1). However, i is the subnumber of the block number in the macroblock MB.
It's a fix. Note that R in the formula (1)_i, F (x)
Are expressed as the following equations (2) and (3), respectively.
It However, << y >> is the maximum integer that does not exceed y. Na
The above formula (1) is based on the theory of D-II, Vo
l. J84_D_II, No. 2,2001.2 "bit rate
Padding for orthogonal transformation of arbitrary shape considering distortion characteristics
Variance and quantization of motion-compensated prediction error
How to estimate the error power code amount from the step size
This is an improvement to be applied to TM5.

On the other hand, the code amount estimation h of the motion vector MV is such that the motion vector of the block i is v _i , and the motion vector of the block (i-1) immediately before the block i is v v.
_{If i− 1} , it can be expressed by the following equation (4). _{h = f (| v i -v} i-1 |) (v when _i ≠ 0) _{(when v i = 0) = 0 ···} (4) This equation (4) is, in MPEG-2, Motion vector MV
Since the variable-length code is applied to the above, it is generally created by paying attention to the fact that the smaller the difference between the motion vectors MV is, the smaller the code amount is between adjacent macroblocks MB.

In step S5, the motion vector MV that minimizes (g + h) is selected. In step S6, i
= N (N is the total number of macroblocks in one image) is determined. When this determination is negative, the process proceeds to step S7 and 1 is added to i. Then, returning to step S2, the type of the next macroblock MB is determined.

On the other hand, if the determination in step S2 is the detail section, the process proceeds to step S8, where S
The vector to the macroblock having the smallest AD is the motion vector.

The present inventor conducted a computer simulation experiment using a test image in order to verify the effectiveness of this embodiment. As the test image, 15 frames of Opening Ceremony (1920 pixels × 1080 lines), which is a standard test image of the Institute of Image Information and Television Engineers, were used. The encoding conditions are a bit rate of 15 Mbit / sec, GOP
The structure used N = 15 and M = 1. As a comparison target, T
An experiment was conducted in M5 under the same conditions as above.

As a result, according to the method of this embodiment, the scattering of the motion vector MV is suppressed in the flat portion of the image such as the track as compared with the case of using TM5, and the skip MB
Was confirmed to have decreased. Therefore, it was found that the method of the present embodiment subjectively suppresses the flickering and the like occurring in the skip MB and the like.

[0021]

As is apparent from the above description, according to the present invention, it is possible to prevent the scattering of the motion vector MV in a flat area of an image or the like. Also, the motion vector MV with the smallest code amount can be detected.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of an encoding device to which the present invention is applied.

FIG. 2 is a flowchart showing the operation of the motion vector detection unit of FIG.

[Explanation of symbols]

1 ... Macroblock extraction unit, 2 ... Subtraction unit, 3 ... Orthogonal transformation unit, 4 ... Quantization unit, 5 ... Variable length coding unit, 10 ...
-Motion compensation unit, 12 ... Motion vector detection unit, 13 ... Macro block type determination unit.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Masahiro Wada             2-15-1 Ohara, Kamifukuoka City, Saitama Stock             Company CAD Research Institute (72) Inventor Shuichi Matsumoto             2-15-1 Ohara, Kamifukuoka City, Saitama Stock             Company CAD Research Institute F-term (reference) 5C059 MA00 MA23 MC11 ME01 NN10                       NN21 NN28 NN38 PP22 PP23                       TD06 TD08 UA02                 5J064 AA01 AA04 BA09 BA16 BB03                       BC08 BC16 BD01

Claims (3)

[Claims] 1. A macroblock type representing the image property of a macroblock of a video signal is determined, and when the macroblock type is a flat region of an image, an estimated amount of error power code amount of the macroblock is used. A motion detection method in a coding apparatus, characterized in that a motion vector having a minimum sum of the motion vector and the estimated code amount is selected. 2. The motion detecting method according to claim 1, wherein the flat area of the image includes at least one of a flat portion and an edge portion of the image. 3. The estimated amount of code amount of the motion vector of the macroblock is a difference from the motion vector of the immediately preceding macroblock.
A motion detection method in the encoding device according to.